Christoph Brockmann wrote:
> Hi everyone,
>
> Using Analysis in the refinement stage of structure calculations, I came
> about some questions/issues/request around the generation and export.
>
> The first one is a question about the "Distance function" in the
> "Generate Distance Constraints" dialogue. Here you can select "NOE bins"
> and "Normalized" but there is only one way do influence the setup (Set
> distance classes). What is the difference between the two methods (I
> cant't find it in the documentation)?
Could find this described in the help pages. is it there?
I'm guessing the 'normalized' option uses some r ~ I^(-1/n) conversion,
with usually n=6 , and some function to give the allowed error margin to
the distance.
Frankly, I don't see the advantage of the bins approach. I think its
a remnant of the times when people counted contours in spectra plotted
on paper.
With bins (if you don't you a great number of them) you lose
information, which you can see from the fact that it not possible
to calculate the origiinal NOE intensities back from the restraint
distances.
The normalized approach should be equivalent to the situation with
infiniteley many bins. So using bins is never better than using some
function r ~ f(I). And with bins you get the situation that very small
calibration changes can put restraints in a different bin.
Having said that, using bins is usually not bad, it's just not optimal
and not easier to use imho. Also it leaves too much room for the user
to fiddle around with bin settings to make violations go away.
>
> The second question is more about the handling of the restraints itself.
> Although some restraints work on pseudoatoms (like methyl groups), this
> fact seems not to be reflected in the distance that should be enlarged
> to account for the difference between the geometric center of a group
> (to which the restraint is applied) and the hydrogens.
If the methyl groups (or other degenerate resonance like flipping ring
protons) are treated with ambiguous restraints there is no need to
enlarge the distance ('pseudo-atom corrections'). The ambiuous restraint
treats the signal as a sum of the individual signals of the protons in
the methyl group. In theory, with perfect data and a perfect structure,
the restraint distance should exactly match the distance (or 1/r^6-sum
distance) in the structure. Like with using bins, with pseudo-atom
corrections you unnecessarily lose information, not so much for methyl
groups, but a lot more for aromatic ring protons.
> Another Issue is
> the handling of non-stereospecificly assigned hydrogens which are
> currently handled as ambiguous restraints. This allows independent
> swapping for each peak/restraint loosing the connection of one resonance
> to a single proton.
I hope that this is not correct. Usually (in e.g. aria) the
stereospecific ambiguity is not treated with ambiguous restraint
(except for treating the 3 protons of a methyl group). The
stereospecific ambiguity is treated by swapping the atoms or coordinates
in the structure calculation, and not independently for every
peak/restraint.
This makes sure that all assignments remain consistent, and you don't
get situations where for one set of peaks the high field NOE is
satisfied by HB2 and in another set of peaks it is satisfied by the HB1
atom.
Using ambiguous restraints in this case here loses information,
although you can get some back by adding restraints corresponding to
the sum of restraints for example, and other combinations.
So if the structure calculation software allows swapping the
coordinates, I think that should be the preferred method.
so I agree with the lines below.
(which I didn't read too well before writing the above, sorry for the
duplication ;-)
> In my understanding this should be handled during
> structure calculation (by the lack of the stereo improper) that allows
> the hydrogens or methyl groups to exchange their positions during the
> high temperature phase of the simulated annealing retaining the
> correspondence of one resonance to one proton.
Eiso
>
> My last point is more of a feature request. It would be nice, if one
> could filter the peaks that are used for restraint generation in one or
> the other way. Things one might want to exclude are intra-residue peaks
> or ones that one "labelled" at some stage as "difficult" either due to
> overlap or because they were close to the diagonal.
>
> thanks,
> Christoph
>
